本研究探討PBL結構下結合生成式人工智慧（AI）工具之多媒體創作任務，於大學西班牙語課程中的實施情形與學習歷程線索。研究場域為一門1學分西語多媒體選修課，課程於期中後導入生成式AI輔助創作專案，要求學生以小組形式完成西語學習導向之多媒體作品，並繳交文法解說表；實作結果中，各組皆選擇歌曲或影片作為產出形式。研究採教學實踐研究取向之混合方法設計，量化資料包括院級總研究計畫統一施測之七構面問卷（數位素養、解決問題與創意思考、批判思考與溝通、小組合作、自主學習、數位公民、跨領域）之前後測描述性趨勢，以及課程評量成績標準化後之相對變化（Δz）作為輔助指標；質性資料則包括作品草稿與定稿比對、文法解說表、同儕互評文字與教師觀察。結果顯示，問卷七構面後測平均數皆高於期初，呈現學生在素養導向能力與任務參與相關面向之正向自我知覺趨勢；然而，Δz呈現相對表現上升、大致穩定與相對表現下降並存之分化結果，顯示課程評量之相對變化不宜被視為語言能力提升之直接證據。質性分析進一步顯示，學生在「草稿—回饋—修訂」歷程中出現語言形式調整與文法重點補強，並透過文法解說表進行語言規則整理與再表述；同儕互評亦反覆指向作品之學習價值與可理解性，如文法重點清楚、可供複習及字幕清楚及語言重點標示明確等。整體而言，本研究之貢獻不在於證明生成式AI能直接提升西語能力，而在於呈現PBL結構下結合AI工具之多媒體任務如何在真實課堂中被實施，並指出作品修訂、文法解說與同儕互評可作為觀察語言加工歷程與素養行為之證據來源。研究建議未來課程可強化語言目標與鷹架設計、提升前後測可比性，並透過多輪行動研究持續改善AI輔助外語任務之設計與評量。
This study examines the implementation of a generative AI–assisted multimedia creation project in a university Spanish course and explores the learning process evidence emerging from this classroom-based intervention. The study was conducted in a one-credit elective course, Multimedia Spanish Learning. After the midterm, students completed an AI-assisted group creation task and submitted a multimedia product together with a grammar explanation sheet; all groups ultimately chose songs or videos as their final products. Adopting a mixed-methods design within a teaching-practice research orientation, the study drew on both quantitative and qualitative data. Quantitative data included descriptive pre/post trends from a seven-dimension survey administered as part of a faculty-wide project—covering digital literacy, problem solving and creative thinking, critical thinking and communication, collaboration, self-directed learning, digital citizenship, and interdisciplinarity—as well as standardized relative change in course performance (Δz) as a supplementary indicator. Qualitative data comprised draft–final comparisons, grammar explanation sheets, peer assessment comments, rubric information, and teacher observations.
The findings show that post-course averages were higher across the seven survey dimensions, suggesting positive descriptive trends in students’ self-perceived competencies and perceptions related to task participation. However, Δz results showed a mixed distribution of relative performance change, indicating that course performance should not be interpreted as direct evidence of language ability gains. Qualitative analyses further revealed patterns of human–AI collaborative revision, including language-form adjustment, reinforcement of grammar focus, and rule explanation through grammar explanation sheets. Peer feedback frequently highlighted learning value and comprehensibility cues, such as clarity of grammar focus and usefulness for review, providing additional process evidence for interpreting the divergent quantitative results. Overall, the contribution of this study lies not in demonstrating that generative AI directly improves Spanish proficiency, but in presenting a feasible PBL-oriented classroom design that integrates AI tools into multimedia language-learning tasks and identifies revision records, grammar explanation sheets, and peer feedback as useful evidence sources for examining language-processing processes and competency-oriented learning behaviors. Implications include strengthening linguistic scaffolding, improving alignment between tasks and assessment, enhancing pre/post comparability, and conducting iterative action cycles to refine AI-supported task design and evaluation.